处理类型异常。

SegFormer/training/dataset.py

@@ -161,10 +161,13 @@ class FoodSegmentationDataset(Dataset):
         # albumentations会自动将增强同时应用到image和mask
         transformed = self.transform(image=image, mask=mask)
         image = transformed['image']  # Tensor (3, H, W)
-        mask = transformed['mask']    # ndarray (H, W)
+        mask = transformed['mask']    # 可能是ndarray或Tensor
         
-        # 3. 将mask转换为Tensor
+        # 3. 将mask转换为Tensor（检查类型）
-        mask = torch.from_numpy(mask).long()
+        if isinstance(mask, np.ndarray):
+            mask = torch.from_numpy(mask).long()
+        else:
+            mask = mask.long()  # 已经是Tensor，直接转换类型
         
         # 4. 检查mask的有效性
         # 确保mask的值在[0, num_classes-1]范围内